Image segmentation method, system, device and medium

Abstract

The invention discloses an image segmentation method, system, device and medium, and relates to the technical field of image segmentation. The method comprises: obtaining an image to be segmented; introducing a heuristic guidance strategy and a differential evolution strategy into a RIME optimization algorithm, initializing the population size and the maximum number of iterations of the optimization algorithm, and randomly initializing the population position; evaluating the fitness values ​​of all individuals in the population, determining the optimal individual and its corresponding fitness; obtaining heuristic guidance information according to the optimal individual and the currently selected individual; updating the population by a soft search method combined with heuristic information, exchanging individuals by a hard puncture strategy, and updating individuals by a differential evolution strategy; obtaining the optimal individual position of the population and its corresponding fitness value after multiple iterative updates, and segmenting the image to be segmented according to the optimal threshold; the method has higher segmentation accuracy and can obtain high-quality segmentation results.

Description

Technical Field

[0001] The present invention relates to the technical field of image segmentation, and in particular to an image segmentation method, system, device and medium. Background Art

[0002] Image segmentation is a key technology in the field of computer vision. Its main purpose is to divide an image into several regions or objects with practical significance. This process usually involves identifying and distinguishing different parts in an image, such as foreground and background, objects and objects, boundaries and regions, etc. Traditional image segmentation methods include threshold-based segmentation and edge detection. Although these methods have achieved certain success in specific application scenarios, they often face problems such as low computational efficiency, poor robustness, and insufficient segmentation accuracy when processing complex images.

[0003] As an effective image segmentation method, multi-threshold image segmentation divides the image into multiple non-overlapping regions by setting multiple thresholds for further analysis and processing. Obtaining high-quality thresholds is crucial to achieving high-quality segmentation results. However, the complexity of threshold selection makes this process challenging. To this end, researchers have gradually turned to using meta-heuristic algorithms to optimize the threshold selection process. Meta-heuristic algorithms have powerful global search capabilities and can find approximately optimal solutions in a larger solution space, becoming the mainstream method for current threshold selection. Today, many meta-heuristic algorithms have achieved good results in threshold optimization. For example, the original RIME algorithm can achieve accurate image segmentation in a shorter time overhead in the task of multi-threshold image segmentation.

[0004] However, when using meta-heuristic algorithms for complex high-dimensional images, the complex details may cause the algorithm to fall into the local optimal trap when searching for the optimal threshold, making it difficult to obtain high-quality image segmentation results. Summary of the invention

[0005] In view of the deficiency of low image segmentation accuracy in the prior art, the present invention proposes an image segmentation method, system, device and medium, introduces a heuristic guidance strategy and a differential evolution strategy into the RIME optimization algorithm, and obtains an optimizer (HDRIME). The image segmentation method constructed by combining HDRIME with a non-local mean two-dimensional histogram and Renyi entropy is more inclined to converge to the global optimum during the search process, and has a higher segmentation accuracy, and can obtain high-quality segmentation results, thereby solving the problems existing in the prior art.

[0006] An image segmentation method comprises the following steps:

[0007] Obtain an image to be segmented;

[0008] The image to be segmented is input into the RIME optimization algorithm, and the population size, individual dimension dim and maximum number of iterations of the RIME optimization algorithm are initialized; the population position is randomly initialized, the fitness values ​​of all individuals in the population are evaluated, and the optimal individual in the current population and its corresponding fitness are determined; wherein the population includes multiple individuals, and each individual represents a set of thresholds of the image to be segmented;

[0009] According to the optimal individual and the currently selected individual, heuristic guidance information is obtained; according to the heuristic guidance information, the population is updated by using a soft search method of the heuristic information, and the dimension of the optimal individual of the updated population is exchanged with the dimension of the currently selected individual by using a hard puncture strategy; the population after the dimension exchange is updated by a differential evolution strategy, wherein each individual is mutated once in each iterative update, and the mutated individual is formed into a new individual by crossing the original individual, the fitness of the new individual is determined, and the optimal individual is selected according to the fitness of the newly generated individual; after multiple iterative updates, when the current number of iterations is greater than or equal to the maximum number of iterations, the position of the optimal individual in the population and its corresponding fitness value are obtained;

[0010] The image to be segmented is segmented according to the optimal threshold.

[0011] Furthermore, before the image to be segmented is input into the RIME optimization algorithm, the image to be segmented is grayed to obtain a corresponding grayscale image, and the image to be segmented is processed using a non-local mean filtering technique to obtain a corresponding non-local mean image, and the grayscale image and the non-local mean image are combined to obtain a 2D histogram for inputting the RIME optimization algorithm.

[0012] Furthermore, the heuristic guidance information is obtained according to the optimal individual and the currently selected individual, which is expressed as:

[0013] HD=k·(X best,j -X i,j )

[0014] Where k represents the influencing factor of heuristic information adjustment, X best,j is the optimal individual in the current iteration, X i,j is the currently selected individual.

[0015] Furthermore, the population is updated by using a soft search method based on heuristic information, and the update process expression is:

[0016]

[0017] in, represents the value of the jth dimension of the newly generated i-th individual, r, rand and h represent random numbers between [0,1], Ub and Lb represent the upper and lower bounds of the value, and E, α, β are variables, respectively expressed as:

[0018]

[0019] Where it represents the current number of iterations, MaxIt represents the maximum number of iterations of the algorithm, and W is a fixed constant.

[0020] Furthermore, the hard puncture strategy is used to exchange the dimensions of the optimal individual of the updated population with the currently selected individual, which is expressed as:

[0021]

[0022] where F normr (Fit i ) represents the normalized value of the current individual’s fitness value.

[0023] Furthermore, the updating of the population after dimension exchange by the differential evolution strategy specifically includes the following steps:

[0024] In each iteration, each individual is mutated once, expressed as:

[0025]

[0026] Among them, U i =[U i,1 U i,2 …U i,D ] represents the individual after mutation, rand is a random number between [0.2, 0.8], p1, p2 and p3 are random numbers between [1, 2, ... N];

[0027] The mutated individual and the original individual are crossed to form a new individual, which is expressed as:

[0028]

[0029] Among them, V i,j Represents the new individual formed after crossover, U i,j represents the individual after mutation, X i,j Indicates the currently selected individual;

[0030] Calculate the fitness of the new individual Select the best fitness, the specific process is expressed as:

[0031]

[0032] The present invention also includes an image segmentation system, comprising:

[0033] An acquisition module, used for acquiring an image to be segmented;

[0034] The initialization module is used to input the image to be segmented into the RIME optimization algorithm, initialize the population size, individual dimension dim and maximum number of iterations of the RIME optimization algorithm; randomly initialize the population position, evaluate the fitness values ​​of all population individuals, and determine the optimal individual in the current population and its corresponding fitness; wherein the population includes multiple individuals, each individual represents a set of thresholds of the image to be segmented;

[0035] The optimal selection module is used to obtain heuristic guidance information based on the optimal individual and the currently selected individual; according to the heuristic guidance information, the population is updated by using the soft search method of the heuristic information, and the dimension of the optimal individual of the updated population is exchanged with the dimension of the currently selected individual by using the hard puncture strategy; the population after the dimension exchange is updated by the differential evolution strategy, wherein each individual is mutated once in each iterative update, and the mutated individual is formed into a new individual by crossing with the original individual, and the fitness of the new individual is determined; the optimal individual is selected according to the fitness of the newly generated individual; after multiple iterative updates, when the current number of iterations is greater than or equal to the maximum number of iterations, the position of the optimal individual in the population and its corresponding fitness value are obtained;

[0036] The segmentation module is used to segment the image to be segmented according to the optimal threshold.

[0037] The present invention also includes an image segmentation computer device, including: a memory, a processor, and a computer program stored in the memory, and the processor implements the steps of the image segmentation method when executing the computer program.

[0038] The present invention also includes a readable storage medium, wherein the readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, they are used to execute the steps of the image segmentation method.

[0039] The present invention provides an image segmentation method, which has the following beneficial effects:

[0040] The present invention introduces a heuristic guidance strategy and a differential evolution strategy into the rime optimization algorithm to obtain an improved rime optimization algorithm, updates the population by adopting a soft search method of heuristic information, and updates the population after dimension exchange by a differential evolution strategy, so that it tends to converge to the global optimum during the search process, and then selects an optimal threshold set, performs image segmentation according to the optimal threshold, has higher segmentation accuracy, and can obtain high-quality segmentation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 is a flowchart of image segmentation in an embodiment of the present invention;

[0042] Figure 2is the original real image in the embodiment of the present invention;

[0043] Figure 3 A grayscale image segmented by the model proposed in the embodiment of the present invention;

[0044] Figure 4 This is a color image segmented by the model proposed in the embodiment of the present invention. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0046] The present invention proposes an image segmentation method, which introduces a heuristic guidance strategy and a differential evolution strategy into the RIME optimization algorithm to obtain an optimizer (HDRIME) with superior performance. The image segmentation method constructed by HDRIME combined with a non-local mean two-dimensional histogram and Renyi entropy is more likely to converge to the global optimum during the search process, and has a higher segmentation accuracy, and can obtain high-quality segmentation results; Figure 1 As shown, the method specifically comprises the following steps:

[0047] S1. Input the image to be segmented and initialize the threshold segmentation level L.

[0048] S2. grayscale the input original image to obtain a corresponding grayscale image, and process the original image using a non-local mean filtering technique to obtain a corresponding non-local mean image.

[0049] S3. Combine the grayscale image and the non-local mean image to obtain a 2D histogram.

[0050] S4. Input the 2D histogram into the algorithm model to calculate the optimal threshold.

[0051] S5, initialization parameters: the population size N of the optimization algorithm, the individual dimension dim, and the maximum number of iterations MaxIt. Specifically, there are N individuals in the algorithm's population, and each individual represents a set of thresholds. Each individual has dim dimensions, and each dimension is a threshold in a set of thresholds. Therefore, the optimal individual points to a set of optimal thresholds.

[0052] S6. Randomly initialize population position: X i (i=1,2,…,N).

[0053] S7. Evaluate the fitness values ​​of all individuals in the population: Renyi entropy.

[0054] S8. Determine the optimal individual X in the current population bestand its corresponding fitness.

[0055] S9, calculate the variables E, α, β, so that the parameters can be adaptively changed according to the operation stage. The calculation scheme is as follows.

[0056]

[0057] Where it represents the current number of iterations, MaxIt represents the maximum number of iterations of the algorithm, and W is a fixed constant, which is set to 5 here.

[0058] S10, calculating the heuristic guidance information, the calculation formula is as follows:

[0059] HD=k·(X best,j -X i,j )(4)

[0060] Where k represents the influencing factor of heuristic information adjustment, which is set to 0.3, and X best,j is the best individual of the contemporary era, X i,j is the currently selected individual.

[0061] S11. Update the population through a soft search method combined with heuristic information, as shown below.

[0062]

[0063] in represents the value of the jth dimension of the newly generated ith individual, r, rand and h represent random numbers between [0,1], and Ub and Lb represent the upper and lower bounds of the value.

[0064] S12. Use the hard puncture strategy to exchange the optimal individual with a certain dimension of the current individual. The execution formula of the strategy is as follows:

[0065]

[0066] in represents the value of the jth dimension of the newly generated i-th individual, F normr (Fit i ) represents the normalized value of the current individual’s fitness value.

[0067] S13. The population obtained after executing the above steps is updated through differential evolution, as follows:

[0068] S13.1. Each individual is mutated once in each iteration, and the formula is as follows:

[0069]

[0070] U i =[Ui,1 U i,2 …U i,D represents the mutated individual, rand is a random number between [0.2, 0.8], and p1, p2, and p3 are random numbers between [1, 2,... N].

[0071] S13.2. Cross the mutated individual and the original individual to form a new individual. The specific operation is shown as follows:

[0072]

[0073] Here, P c is set to 0.3.

[0074] S13.3. Calculate the fitness of the new individual and select the best one. The specific operation is as shown in Equation (9).

[0075]

[0076] Among them, is the newly generated individual.

[0077] S14. Determine whether the iteration termination condition it < MaxIt is satisfied. If it is satisfied, return to step 5 to continue the iteration; otherwise, terminate the iteration.

[0078] S15. Return the optimal threshold and the maximum Renyi entropy: the position of the optimal individual in the population and its corresponding fitness value;

[0079] S16. Perform image segmentation based on the obtained optimal threshold.

[0080] Generally speaking, the present invention uses the Renyi entropy as the objective function, maximizes the Renyi entropy by using the improved RIME algorithm, and the obtained threshold set is the optimal threshold. Combine the grayscale image corresponding to the original image and the non-local mean image to construct a non-local mean two-dimensional histogram to represent the image information.

[0081] In the above two-dimensional histogram, the basic idea of the non-local mean filtering method is that the new predicted value of the requested pixel is equal to the weighted average of all pixels in the entire image that have a similar local structure to it. Assume that I(i) is the grayscale value of pixel i in image I, and I(j) is the grayscale value of pixel j. Then the non-local mean of image I is calculated as follows:

[0082]

[0083] Where O(p) is the non-local mean filter value of pixel p, ω(p, q) represents the weight of pixel p and pixel q, σ is the standard deviation, μ(p) and μ(q) are local means, L(p) and L(q) are image blocks centered on pixel p and pixel q with a size of m×m. Based on the above method, the grayscale images are combined to form a non-local mean image, and then a two-dimensional histogram is formed. If the number of layers of the original grayscale image O(x, y) is [0, L-1] and the image size is M×N, then the corresponding non-local mean image G(x, y) also has the same number of layers and size. On this basis, a plane is created using the combination of O(x, y) and G(x, y), where G(x, y) represents the frequency of occurrence of the corresponding pixel combination G(x, y), i represents the grayscale value of pixel i in O(x, y), and j represents the grayscale value of pixel j in G(x, y). Therefore, the final two-dimensional histogram is normalized by formula (14) to obtain

[0084]

[0085] L-1 gray levels can be expressed as:

[0086] {T1, T2, ..., T L-1}={(s1,t1),(S2,t2),...,(S L-1 , t L-1 ))}

[0087] where {t1, t2, ..., L-1} represents the value of the grayscale image, and {s1, s2, ..., L-1} represents the value of the non-local mean image. The corresponding image can be divided into L-1 categories {C1, C2, ..., C L-1}. Calculate Renyi entropy R based on the image information expressed by the two-dimensional histogram α , defined as formula (15):

[0088]

[0089] Therefore, if Renyi entropy is used as the objective function, then HDRIME finds the value that makes R α The maximized threshold set is the optimal segmentation threshold.

[0090] against Figure 2 The original real image (color) in , the segmentation results of the proposed model are as follows Figure 3 and Figure 4In order to make the argument more rigorous, the present invention uses other multi-threshold image segmentation models to compare with the proposed model, including three original algorithms: Harris Hawk Algorithm (HHO), Whale Optimization Algorithm (WOA), Sin-Cosine Optimization Algorithm (SCA), and three well-known improved algorithms (IWOA, B LPSO, SCADE).

[0091] Tables 1 to 3 show the analysis results of feature similarity (FSIM), peak signal-to-noise ratio (PSNR), and structural similarity (SSIM) using Wilcoxon signed rank test and Friedman signed rank test. Among them, 'R+' indicates that the multi-threshold image segmentation method based on HDRIME of the present invention is better than other similar methods, 'R-' indicates that the multi-threshold image segmentation method based on HDRIME of the present invention is worse than other similar methods, and 'R=' indicates that the multi-threshold image segmentation method based on HDRIME of the present invention has similar segmentation performance to other similar methods. 'ARV' indicates the overall ranking mean level of each method, and 'Rank' indicates the ranking level of each method.

[0092] Table 1 Analysis results of Wilcoxon signed rank test and Friedman signed rank test for feature similarity

[0093]

[0094] Table 2 Analysis results of Wilcoxon signed rank test and Friedman signed rank test for peak signal-to-noise ratio

[0095]

[0096]

[0097] Table 3 Analysis results of Wilcoxon signed rank test and Friedman signed rank test for structural similarity

[0098]

[0099] Through the analysis results of Wilcoxon signed rank test and Friedman signed rank test, the multi-threshold image segmentation method based on HDRIME of the present invention can not only obtain excellent segmentation results at low thresholds (2, 4, 6), but also maintain good performance at high thresholds (15, 18, 20), and almost achieve the first place at all thresholds. This fully demonstrates that the multi-threshold image segmentation method based on HDRIME is a multi-threshold image segmentation method that can obtain high-quality segmentation results and is very stable.

[0100] The present invention proposes a multi-threshold image segmentation model based on HDRIME. The model is based on non-local mean two-dimensional histogram and Renyi entropy, and introduces heuristic guidance strategy and differential evolution strategy into the rime optimization algorithm to obtain an improved rime optimization algorithm (HDRIME), so as to select the optimal threshold set, thereby obtaining a higher quality segmentation result.

[0101] Based on the same inventive concept, the present invention also proposes an image segmentation system, comprising:

[0102] The acquisition module is used to acquire the image to be segmented.

[0103] The initialization module is used to input the image to be segmented into the RIME optimization algorithm, initialize the population size, individual dimension dim and maximum number of iterations of the RIME optimization algorithm; randomly initialize the population position, evaluate the fitness values ​​of all population individuals, and determine the optimal individual in the current population and its corresponding fitness; wherein the population includes multiple individuals, each individual represents a set of thresholds for the image to be segmented.

[0104] The optimal selection module is used to obtain heuristic guidance information based on the optimal individual and the currently selected individual; based on the heuristic guidance information, the population is updated by using the soft search method of the heuristic information, and the dimension of the optimal individual of the updated population is exchanged with the dimension of the currently selected individual by using the hard puncture strategy; the population after the dimension exchange is updated by the differential evolution strategy, wherein each individual is mutated once in each iterative update, and the mutated individual is formed into a new individual by crossing the original individual, and the fitness of the new individual is determined; the optimal individual is selected according to the fitness of the newly generated individual; after multiple iterative updates, when the current number of iterations is greater than or equal to the maximum number of iterations, the position of the optimal individual in the population and its corresponding fitness value are obtained.

[0105] The segmentation module is used to segment the image to be segmented according to the optimal threshold.

[0106] The present invention also provides an image segmentation computer device, comprising: a memory, a processor and a computer program stored in the memory, and the processor implements the steps of the image segmentation method when executing the computer program.

[0107] The present invention further provides a readable storage medium, wherein the readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, the steps of the image segmentation method are executed.

[0108] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. An image segmentation method, characterized in that: The following steps are involved: Obtain an image to be segmented; The image to be segmented is input into the Rime Optimization Algorithm RIME, and the population size, individual dimension and maximum number of iterations of RIME are initialized; the population position is randomly initialized, the fitness values ​​of all individuals in the population are evaluated, and the optimal individual in the current population and its corresponding fitness are determined; The population includes multiple individuals, each of which represents a set of thresholds for the image to be segmented; According to the optimal individual and the currently selected individual, heuristic guidance information is obtained; According to the heuristic guidance information, the population is updated by using the soft search method of the heuristic information, and the dimension of the optimal individual of the updated population is exchanged with the dimension of the currently selected individual by using the hard puncture strategy; the population after the dimension exchange is updated by the differential evolution strategy, in which each individual is mutated once in each iterative update, and the mutated individual is formed into a new individual by crossing with the original individual, and the fitness of the new individual is determined; the optimal individual is selected according to the fitness of the newly generated individual; after multiple iterative updates, when the current number of iterations is greater than or equal to the maximum number of iterations, the position of the optimal individual in the population and its corresponding fitness value are obtained; The image to be segmented is segmented according to the optimal threshold.

2. An image segmentation method according to claim 1, characterized in that: Before the image to be segmented is input into the rime optimization algorithm RIME, the image to be segmented is grayed to obtain a corresponding gray image, and the image to be segmented is processed using a non-local mean filtering technique to obtain a corresponding non-local mean image, and the gray image and the non-local mean image are combined to obtain a 2D histogram for inputting the RIME optimization algorithm.

3. The image segmentation method according to claim 1, characterized in that: According to the optimal individual and the currently selected individual, the heuristic guidance information is obtained, which is expressed as: HD=k·(X best,j -X i,j ) Where k represents the influencing factor of heuristic information adjustment, X best,j is the optimal individual in the current iteration, X i,j is the currently selected individual.

4. The image segmentation method according to claim 3, characterized in that: The population is updated by using a soft search method using heuristic information, and the update process expression is: in, represents the value of the jth dimension of the newly generated i-th individual, r, rand and h represent random numbers between [0,1], Ub and Lb represent the upper and lower bounds of the value, and E, α, β are variables, respectively expressed as: Where it represents the current number of iterations, MaxIt represents the maximum number of iterations of the algorithm, and W is a fixed constant.

5. An image segmentation method according to claim 4, characterized in that: The hard puncture strategy is used to exchange the dimensions of the optimal individual of the updated population with the currently selected individual, which is expressed as: where F normr (Fit i ) represents the normalized value of the current individual’s fitness value.

6. The image segmentation method according to claim 1, characterized in that: The method of updating the population after dimension exchange by using the differential evolution strategy specifically includes the following steps: In each iteration, each individual is mutated once, expressed as: Among them, U i =[U i,1 U i,2 …U i,D ] represents the individual after mutation, rand is a random number between [0.2, 0.8], p1, p2 and p3 are random numbers between [1, 2, ... N]; The mutated individual and the original individual are crossed to form a new individual, which is expressed as: Among them, V i,j Represents the new individual formed after crossover, U i,j represents the individual after mutation, X i,j Indicates the currently selected individual; Calculate the fitness of the new individual Select the best fitness, the specific process is expressed as:

7. An image segmentation system, characterized in that: include: An acquisition module, used for acquiring an image to be segmented; The initialization module is used to input the image to be segmented into the RIME optimization algorithm, initialize the population size, individual dimension and maximum number of iterations of the RIME optimization algorithm; randomly initialize the population position, evaluate the fitness values ​​of all population individuals, and determine the optimal individual in the current population and its corresponding fitness; The population includes multiple individuals, each of which represents a set of thresholds for the image to be segmented; The optimal selection module is used to obtain heuristic guidance information according to the optimal individual and the currently selected individual; According to the heuristic guidance information, the population is updated by using the soft search method of the heuristic information, and the dimension of the optimal individual of the updated population is exchanged with the dimension of the currently selected individual by using the hard puncture strategy; the population after the dimension exchange is updated by the differential evolution strategy, in which each individual is mutated once in each iterative update, and the mutated individual is formed into a new individual by crossing with the original individual, and the fitness of the new individual is determined; the optimal individual is selected according to the fitness of the newly generated individual; after multiple iterative updates, when the current number of iterations is greater than or equal to the maximum number of iterations, the position of the optimal individual in the population and its corresponding fitness value are obtained; The segmentation module is used to segment the image to be segmented according to the optimal threshold.

8. An image segmentation computer device, characterized in that: include: A memory, a processor, and a computer program stored in the memory, wherein the processor implements the steps of the image segmentation method according to any one of claims 1 to 6 when executing the computer program.

9. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and the computer program includes program instructions. When the program instructions are executed by a processor, they are used to execute the steps of the image segmentation method according to any one of claims 1 to 6.

Images